Comparison of Short-Term Electrical Load Forecasting Models using Datasets from The Building Automation System in The Department of Electrical Engineering ITN

Overloading in electrical systems can cause hazards such as damaging electrical devices, melting cable lines, and potentially even a fire. Monitoring the use of electric load alone is less effective in preventing overload. The short-term forecasting process is needed for the use of the load. In this paper, we describe a comparison of electrical load forecasting methods consisting of the Kalman Filter, Autoregressive Integrated Moving Average (ARIMA), and Long Short-Term Memory (LSTM) Network. The dataset was used in the form of a reading history of power sensors installed in the Building Automation System (BAS) of Institut Teknologi Nasional (ITN) Malang. The experiment was carried out by comparing the loss rate in each forecasting model with the Root Mean Square Error (RMSE) calculation method. The results obtained include an average loss value of29.474 for forecasting electrical load using Kalman Filter, another average loss value of 29.136 for forecasting electrical load using ARIMA, another average loss value of27.931 for forecasting electrical load using LSTM with three input variables, and another average loss value of 28.049 for forecasting electrical loads using LSTM with six input variables. These results indicate that the LSTM model with three input variables has the smallest average loss level compared to other models.